Repository navigation
docs(atlas): Workforce is the product; Conductor is a team config - #1635
Conversation
Supersede the sibling-track / later-triangulation lock. Teach Workforce as the L2 product surface and Conductor as one team whose seats include coding-harness workers. Mark omp, seat wiring, and OnSecurity as proposed; park the #1612 console half; keep FIX-1320 and #1600 as floor. Co-authored-by: Jake Hoffner <jhoffner@users.noreply.github.com>
|
The latest updates on your projects. Learn more about Vercel for GitHub. 3 Skipped Deployments
|
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
There was a problem hiding this comment.
Simplify review — direction sound, repetition is the cost
The product lock is right: Workforce is the shippable Layer 2 product; Conductor is one team config on it. Cross-links to workforce.html#conductor and the conductor/framework deferrals are the correct architecture for this atlas set.
The concern is not the lock — it is how much surface area the PR rewrites to say the same sentence. Net diff is ~179 insertions / ~178 deletions: complexity moved from “two sibling products” to many parallel copies of one claim across mastheads, ledes, SVG labels, tables, gates, cuts, and the one-sentence recap. Performance impact is negligible (~3 KB on workforce.html); the cost is reader and maintainer load.
Cleaner approach (if revisiting)
A surgical pass could have shipped the same lock with less churn:
- Canonize in
workforce.html§15 only — masthead + §15 + (maybe) one diagram. - Pointer sentences elsewhere — conductor/framework/index/README get one line + link, not restated pitch.
- Bundle scope separately —
omp, OnSecurity GTM naming, and #1612 console park are orthogonal; they belong in §15 + §20 cuts, not sprinkled through §01 SVGs, §16 inventory, §19 gates, and conductor thread bodies. - One archaeology note — “supersedes separate-tracks / later-triangulation” once in §15; drop repeated negation in figcaptions and sib lines.
- Match README on
index.html— README warns roadmap is stale; index still sells “three jobs” with no disclaimer.
What I would not block on
- Leaving
roadmap.htmlfor Strategy (README disclaimer is enough if index matches). - Replacing the Eve/Grok analog — tradeoff, not clearly wrong.
- Atlas self-contained HTML duplication (pre-existing pattern; extracting shared CSS is a separate refactor).
Skipped / your call
- Whether heavy “superseded” negation is deliberate archaeology for readers of the prior pin.
- Whether
omp/ OnSecurity prominence reflects active GTM you want repeated everywhere.
Verdict: COMMENT — mergeable direction; a dedupe pass could cut ~30–40% of the new prose without weakening honesty tags. Inline notes on the highest-leverage spots.
Sent by Cursor Automation: Review For Best Approach
| <h1>One worker contract. Then compose.</h1> | ||
| <p class="standfirst">Workforce is the general Layer 2 convention: <strong>one worker shape, one flow per worker</strong>. A <strong>seat</strong> is a roster slot that binds a <strong>worker</strong> — not an Agent. A worker is any flow on that contract; thin sequencers count. <strong>Agent</strong> (product) is the opinionated worker: sequencer + generators + persona + durable member memory. A seat can be any worker. Thin seats do not need <code>AgentRegistry</code> or member memory. The target is the <strong>RC collaboration story — teams, plural</strong>. The build-out is <a href="#vision">§03</a>. The three words are <a href="#fence">§04</a>. This page teaches the target. It does not ship it. Conductor does not wait on it.</p> | ||
| <h1>Workforce is the product. Conductor is a team on it.</h1> | ||
| <p class="standfirst">Workforce is the <strong>Layer 2 product</strong>: roster / teams + seats + rooms + thin helpers. It can ship for a client (OnSecurity is a GTM name, not an impl) and for personal or org workforces. A <strong>seat</strong> binds a <strong>worker</strong>, not an Agent. A worker is any flow on the contract; thin sequencers count. <strong>Agent</strong> (product) is the opinionated worker: sequencer + generators + persona + durable member memory. <strong>Conductor</strong> is one consumer: a Workforce team whose seats include coding-harness workers. The roster <em>names</em> the harness; <strong>omp</strong> (opinionated Pi harness) can fill that seat — proposed, not integrated. One harness-manager stack. Layer 1 stays substrate. The three words are <a href="#fence">§04</a>. The build-out is <a href="#vision">§03</a>. Astra gaps stay in <a href="#gaps">§18</a>.</p> |
There was a problem hiding this comment.
Scope bundling. The standfirst now carries the full product lock plus OnSecurity GTM, omp, harness-manager, and Conductor-as-consumer in one paragraph. Consider keeping the masthead to Workforce vocabulary (seat/worker/agent/teams) and deferring Conductor/omp/GTM to §15 — otherwise every reader hits three orthogonal decisions before §01.
| <span><i class="sw" style="background: var(--prose-soft); border: 1.2px solid var(--prose)"></i> later / cut — not foundation</span> | ||
| </div> | ||
| <p class="sib">Pinned to <code>origin/main</code> at <code>d8899c9b5</code> (2026-09-07, #1614). Sibling: <a href="conductor.html">Conductor</a> — this-cycle coding product on the same dispatch spine. Cross-flow PR <a href="https://github.com/fixpoint-labs/flow-state-dev/pull/1600">#1600</a> is open; the refuse is still on this pin.</p> | ||
| <p class="sib">Pinned to <code>origin/main</code> at <code>088181af7</code> (2026-09-07). Conductor is a Workforce config, not a sibling product and not a parallel Proof. The old lock — separate tracks, later triangulation — is <strong>superseded</strong>. Cross-flow PR <a href="https://github.com/fixpoint-labs/flow-state-dev/pull/1600">#1600</a> is open; instances (<a href="https://linear.app/fixpoint-labs/issue/FIX-1320">FIX-1320</a>) still first-wins. GTM timing stays Strategy.</p> |
There was a problem hiding this comment.
Superseded archaeology. “The old lock … is superseded” plus sibling-ship negation repeats in figcaptions, §15, gates, and cuts. One short “what changed” note in §15 should be enough; readers who never saw the old lock pay rent on every section.
| <div class="sec-head"><span class="sec-num">15</span><h2>Conductor</h2></div> | ||
| <p class="lede">This cycle, Conductor is a <strong>thin shell on Layer 1 dispatch</strong>. It does not wait on a Workforce package, including a foundation kick. It <em>does</em> care about slice 1 of the build-out (cross-flow, resume, bare-kind, member memory). Conductor-as-a-coding-extension-on-Workforce is slice 4, not a dependency arrow. If a team seat names a harness, it rides Conductor’s harness-manager track — one stack; team/roster only names the harness as L2 config — once that path exists. The two products stay triangulated. GTM timing is Strategy.</p> | ||
| <div class="sec-head"><span class="sec-num">15</span><h2>Conductor as a Workforce config</h2></div> | ||
| <p class="lede">The lock: <strong>Workforce is the product. Conductor is one Workforce team</strong> whose seats include coding-harness workers. The roster names the harness. <code>omp</code> can fill that seat for now — proposed, not integrated. One harness-manager stack. Do not grow a Conductor-only path beside Workforce. The old lock — separate tracks, later triangulation, Conductor-as-L3-on-Workforce as a horizon — is <strong>superseded</strong>. GTM timing is Strategy.</p> |
There was a problem hiding this comment.
Canonical home — good. This §15 lede is the right place to own the lock, exists-vs-proposed split, and omp/harness-manager honesty. I'd treat this as the only place that states the full pitch; other sections could link here instead of restating (masthead h1 + link is enough upstream).
| <div class="note" style="margin-top:26px"> | ||
| <span class="lbl">The one-sentence version</span> | ||
| <p>Workforce is one worker contract on Layer 1 — a sequencer flow; generator calls do not mint a second type. A seat is a roster slot that binds a worker, not an Agent. Thin seats skip <code>AgentRegistry</code> and member memory. Product Agent is L2 opinion (sequencer + generators + persona + member memory). The core <code>Agent</code> / <code>AgentRegistry</code> types are the thin seam; do not grow them; do not delete them; engine never sees the registry as product. The target is RC collaboration: teams, plural. Build-out is 1→4. Readiness is named gaps on that inventory — instances, then conversation + board proofs, then durable delegation, then the materializer. A roster is a seat-list. A team is that roster plus shared rooms plus dispatch — Layer 2, not an L1 <code>Team</code> type, not a TeamFlow. Product folders are <code>teams/<teamId>/workers/<name>/</code>. Projects live as resources, boards, sessions, and content — no <code>projects/</code> tree, no <code>Project</code> type. A new seat is a file on that path. Scan feeds today’s <code>defineFlow</code> for every seat; <code>AgentRegistry</code> only when that seat is an opinionated agent. A tree scan is a named gap. Hot-dynamic kinds after boot are a named gap. Address-the-team is an intake seat, a worker with a static DM. Identity lives on agent/member scope, not on the seat. Knowledge is four uses of existing <code>user</code> / <code>org</code> resources. Member identity is the named gap — prove <code>flowIsolation</code> first. Helpers take a team key from day one. Ordered multi-reply is proposed Layer 2. RC gates are proposed until they are true. No second Agent. No second registry. No one-team-only API. Cross-flow is still refused on this pin. Conductor does not wait. GTM is Strategy.</p> | ||
| <p>Workforce is the Layer 2 product: one worker contract on Layer 1 — a sequencer flow; generator calls do not mint a second type. A seat is a roster slot that binds a worker, not an Agent. Thin seats skip <code>AgentRegistry</code> and member memory. Product Agent is L2 opinion (sequencer + generators + persona + member memory). The core <code>Agent</code> / <code>AgentRegistry</code> types are the thin seam; do not grow them; do not delete them; do not fold Workforce into core; engine never sees the registry as product. Conductor is one Workforce team whose seats include coding-harness workers. The roster names the harness. <code>omp</code> can fill that seat and is proposed, not integrated. One harness-manager stack. Park Conductor-only surfaces that skip Workforce (#1612 console half). The target is RC collaboration: teams, plural. Build-out is 1→4. Readiness is named gaps on that inventory — instances, then conversation + board proofs, then durable delegation, then the materializer. A roster is a seat-list. A team is that roster plus shared rooms plus dispatch — Layer 2, not an L1 <code>Team</code> type, not a TeamFlow. Product folders are <code>teams/<teamId>/workers/<name>/</code>. Projects live as resources, boards, sessions, and content — no <code>projects/</code> tree, no <code>Project</code> type. A new seat is a file on that path. Scan feeds today’s <code>defineFlow</code> for every seat; <code>AgentRegistry</code> only when that seat is an opinionated agent. A tree scan is a named gap. Hot-dynamic kinds after boot are a named gap. Address-the-team is an intake seat, a worker with a static DM. Identity lives on agent/member scope, not on the seat. Knowledge is four uses of existing <code>user</code> / <code>org</code> resources. Member identity is the named gap — prove <code>flowIsolation</code> first. Helpers take a team key from day one. Ordered multi-reply is proposed Layer 2. RC gates are proposed until they are true. No second Agent. No second registry. No one-team-only API. Cross-flow is still refused on this pin. Multi-hire waits on FIX-1320. GTM is Strategy.</p> |
There was a problem hiding this comment.
| </div> | ||
| <p class="sib">Two names collide on this page, and every use keeps them apart. The board's <code>dispatcher: "topological"</code> is an ordering <em>strategy</em> (exists); the <code>dispatcher()</code> of section 6 is a block <em>factory</em>, and it exists. This page's <code>inbox/**</code> is the <em>question collection</em>; the dispatch-protocol design's “one inbox per flow” is the <em>typed entry set</em>.</p> | ||
| <p class="sib">Sibling: <a href="workforce.html">Workforce atlas</a> — the Layer 2 convention. This cycle writes that page. Conductor does not wait on a Workforce product.</p> | ||
| <p class="sib">Workforce is the Layer 2 product. Conductor is <strong>one Workforce team config</strong> whose seats include coding-harness workers — not a sibling ship, not a parallel Proof. The lock lives on the <a href="workforce.html#conductor">Workforce atlas §15</a>. This page stays the Conductor encyclopedia; do not grow Conductor-only surfaces that skip Workforce (park the #1612 console half).</p> |
There was a problem hiding this comment.
Good deferral shape. Short pointer to Workforce §15 + encyclopedia boundary is exactly what conductor.html should do. If you dedupe workforce.html, keep this as the template for framework/index too.
|
|
||
| <div class="tension"> | ||
| <h5>11 · Which layer of workforce do Conductor’s roles sit on?</h5> | ||
| <p><strong>Product fork closed.</strong> Workforce is the product; Conductor is a team config on it. The remaining question is mechanical — where <code>materializeAgent</code> lives — not whether Conductor stays a sibling ship. See <a href="workforce.html#conductor">Workforce atlas §15</a>.</p> |
There was a problem hiding this comment.
Thread 11 framing drift. “Product fork closed” is right, but the heading still asks “Which layer of workforce…” and the next paragraphs reopen a “fork” between wrapper vs primitives. Consider retitling to the mechanical question (materializeAgent placement) and trimming wrapper/primitives prose that reads like an unresolved product fork.
| <div class="sec-head"><span class="sec-num">04</span><h2>Workforce: the layer that isn't built</h2></div> | ||
| <p class="lede">Workforce is meant to be the top of the stack — teams of named agents that work together, powered by every system beneath them. Today it is 397 lines that define <em>what one agent is</em>, and nothing that defines <em>how a set of them relate</em>.</p> | ||
| <div class="sec-head"><span class="sec-num">04</span><h2>Workforce: the Layer 2 product, not a Layer 1 type</h2></div> | ||
| <p class="lede">Workforce is the shippable Layer 2 product — roster / teams + seats + rooms + thin helpers. Conductor is one team config on it, not a sibling ship. Today the package is still 397 lines that define <em>what one agent is</em>, and nothing that defines <em>how a set of them relate</em>. The product lock and the honesty tags live on the <a href="workforce.html">Workforce atlas</a>. Do not mint Team / Channel / MessageBoard here, and do not fold Workforce into core.</p> |
There was a problem hiding this comment.
Duplicate pitch. §04 lede now restates the Workforce product lock (Conductor-as-config, no sibling ship) that lives on the Workforce atlas. Framework inventory should stay L1/L2 boundary + link to workforce.html#conductor, not a second copy of Strategy positioning.
| board plus one session per subscriber, talk on the dispatch spine. | ||
| Conductor is a team config on this surface. The page teaches the lock and | ||
| the honesty tags; the factory is still proposed. | ||
| - `roadmap.html` — the public product horizon. Still written in the older |
There was a problem hiding this comment.
Roadmap stale note — good. Explicit that roadmap.html still uses sibling framing is the right honesty tag. index.html roadmap blurb wasn't updated to match; easy consistency fix if you want one front door.


The atlas still taught Jake’s old lock: Workforce and Conductor as sibling tracks, Conductor-as-L3-on-Workforce as a later triangulation. That sentence is now wrong. A reader could keep shipping Conductor-only surface, or wait to “put Conductor on Workforce later,” as if Workforce were not the product.
This rewrite teaches the room lock: Workforce is the Layer 2 product. Conductor is one Workforce team whose seats include coding-harness workers. The roster names the harness.
ompcan fill that seat and is marked proposed. One harness-manager stack. Layer 1 stays substrate.#1612console half is a parked Conductor-only cut.What's asked of you
Nothing here is yours to decide. This follows the room lock (Jake / Cycle PM / Architect / Systems Thinker).
Parts worth reviewing closely
The rest of this is for whoever reviews the pages.
Linear: room lock (Jake / Cycle PM / Architect / Systems Thinker). Related FIX-1320, #1600, #1612. Atlas-only. No changeset needed.
roadmap.htmlstill uses the older three-jobs / sibling framing; Strategy owns that rewrite.How to review this — altitude, what's in scope, what's deliberately unsettled
This is an atlas editorial following a room lock. No spec, no package change. In scope: whether the pages now teach Workforce as the product and Conductor as a config, whether exists vs proposed vs cut matches main, and whether anything claimed as existing was invented. Out of scope: implementing the factory, landing #1600, integrating
omp, rewriting the Conductor encyclopedia, rewritingroadmap.html.Engineering calls — decided along the way, no product sign-off needed
088181af7(2026-09-07), currentorigin/main.ompis not in the tree; tagged proposed. Claude Code and Codex already fillharness-manager.#1612PI-1..3 (operator console) named as park/cut because it does not ride Workforce. The harness half of that epic is not declared dead.roadmap.htmlleft alone on purpose. README now says it is the older framing and does not outrank this lock.materializeAgentquestion is left as implementer.Everything else
What landed
workforce.html: new standfirst and product/config diagrams; §15 rewritten; §16 / §19 / §20 honesty; Astra / seat / worker / agent / 1→4 build-out kept.conductor.html/framework.html: honesty pointers only.index.html/README.md: listing copy.Verification: HTML parses (empty stack, no mismatch) on workforce, conductor, framework, index. Twenty workforce sections. Cross-flow not tagged exists.
omp/ OnSecurity not claimed shipped.Atlas URL after merge:
https://fixpoint-labs.github.io/flow-state-dev/workforce.html(GitHub Pages root isdocs/atlas). Until then, the PR copy ofdocs/atlas/workforce.html.